Chromium Code Reviews

Side by Side Diff: src/opts/SkXfermode_opts_SSE2.cpp

Issue 234653002: Xfermode: SSE2 implementation of darken&lighten mode (Closed) Base URL: https://skia.googlesource.com/skia.git@master
Patch Set: Created 6 years, 8 months ago
Use n/p to move between diff chunks; N/P to move between comments.
Jump to:
View unified diff |
« no previous file with comments | « no previous file | no next file » | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 #include "SkColorPriv.h" 1 #include "SkColorPriv.h"
2 #include "SkColor_opts_SSE2.h" 2 #include "SkColor_opts_SSE2.h"
3 #include "SkMathPriv.h" 3 #include "SkMathPriv.h"
4 #include "SkXfermode.h" 4 #include "SkXfermode.h"
5 #include "SkXfermode_opts_SSE2.h" 5 #include "SkXfermode_opts_SSE2.h"
6 #include "SkXfermode_proccoeff.h" 6 #include "SkXfermode_proccoeff.h"
7 7
8 //////////////////////////////////////////////////////////////////////////////// 8 ////////////////////////////////////////////////////////////////////////////////
9 // 4 pixels SSE2 version functions 9 // 4 pixels SSE2 version functions
10 //////////////////////////////////////////////////////////////////////////////// 10 ////////////////////////////////////////////////////////////////////////////////
(...skipping 66 matching lines...)
77 __m128i g = blendfunc_multiply_byte_SSE2(sg, dg, sa, da); 77 __m128i g = blendfunc_multiply_byte_SSE2(sg, dg, sa, da);
78 78
79 79
80 __m128i sb = SkGetPackedB32_SSE2(src); 80 __m128i sb = SkGetPackedB32_SSE2(src);
81 __m128i db = SkGetPackedB32_SSE2(dst); 81 __m128i db = SkGetPackedB32_SSE2(dst);
82 __m128i b = blendfunc_multiply_byte_SSE2(sb, db, sa, da); 82 __m128i b = blendfunc_multiply_byte_SSE2(sb, db, sa, da);
83 83
84 return SkPackARGB32_SSE2(a, r, g, b); 84 return SkPackARGB32_SSE2(a, r, g, b);
85 } 85 }
86 86
87 static inline __m128i darken_byte_SSE2(const __m128i& sc, const __m128i& dc,
88 const __m128i& sa, const __m128i& da) {
89 __m128i sd = _mm_mullo_epi16(sc, da);
90 __m128i ds = _mm_mullo_epi16(dc, sa);
91
92 __m128i cmp = _mm_cmplt_epi32(sd, ds);
93
94 __m128i ret1 = _mm_add_epi32(sc, dc);
mtklein 2014/04/24 17:55:12 I'm not sure why the original code was written so
qiankun 2014/04/25 14:13:24 Done.
95 ret1 = _mm_sub_epi32(ret1, SkDiv255Round_SSE2(ds));
96
97 __m128i ret2 = _mm_add_epi32(dc, sc);
98 ret2 = _mm_sub_epi32(ret2, SkDiv255Round_SSE2(sd));
99
100 __m128i ret = _mm_or_si128(_mm_and_si128(cmp, ret1),
101 _mm_andnot_si128(cmp, ret2));
102
103 return ret;
104 }
105
106 static __m128i darken_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
107 __m128i sa = SkGetPackedA32_SSE2(src);
108 __m128i da = SkGetPackedA32_SSE2(dst);
109
110 __m128i a = srcover_byte_SSE2(sa, da);
111 __m128i r = darken_byte_SSE2(SkGetPackedR32_SSE2(src),
112 SkGetPackedR32_SSE2(dst), sa, da);
113 __m128i g = darken_byte_SSE2(SkGetPackedG32_SSE2(src),
114 SkGetPackedG32_SSE2(dst), sa, da);
115 __m128i b = darken_byte_SSE2(SkGetPackedB32_SSE2(src),
116 SkGetPackedB32_SSE2(dst), sa, da);
117 return SkPackARGB32_SSE2(a, r, g, b);
118 }
119
120 static inline __m128i lighten_byte_SSE2(const __m128i& sc, const __m128i& dc,
121 const __m128i& sa, const __m128i& da) {
122 __m128i sd = _mm_mullo_epi16(sc, da);
123 __m128i ds = _mm_mullo_epi16(dc, sa);
124
125 __m128i cmp = _mm_cmpgt_epi32(sd, ds);
126
127 __m128i ret1 = _mm_add_epi32(sc, dc);
mtklein 2014/04/24 17:55:12 Same here.
qiankun 2014/04/25 14:13:24 Done.
128 ret1 = _mm_sub_epi32(ret1, SkDiv255Round_SSE2(ds));
129
130 __m128i ret2 = _mm_add_epi32(dc, sc);
131 ret2 = _mm_sub_epi32(ret2, SkDiv255Round_SSE2(sd));
132
133 __m128i ret = _mm_or_si128(_mm_and_si128(cmp, ret1),
134 _mm_andnot_si128(cmp, ret2));
135
136 return ret;
137 }
138
139 static __m128i lighten_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
140 __m128i sa = SkGetPackedA32_SSE2(src);
141 __m128i da = SkGetPackedA32_SSE2(dst);
142
143 __m128i a = srcover_byte_SSE2(sa, da);
144 __m128i r = lighten_byte_SSE2(SkGetPackedR32_SSE2(src),
145 SkGetPackedR32_SSE2(dst), sa, da);
146 __m128i g = lighten_byte_SSE2(SkGetPackedG32_SSE2(src),
147 SkGetPackedG32_SSE2(dst), sa, da);
148 __m128i b = lighten_byte_SSE2(SkGetPackedB32_SSE2(src),
149 SkGetPackedB32_SSE2(dst), sa, da);
150 return SkPackARGB32_SSE2(a, r, g, b);
151 }
152
87 //////////////////////////////////////////////////////////////////////////////// 153 ////////////////////////////////////////////////////////////////////////////////
88 154
89 typedef __m128i (*SkXfermodeProcSIMD)(const __m128i& src, const __m128i& dst); 155 typedef __m128i (*SkXfermodeProcSIMD)(const __m128i& src, const __m128i& dst);
90 156
91 extern SkXfermodeProcSIMD gSSE2XfermodeProcs[]; 157 extern SkXfermodeProcSIMD gSSE2XfermodeProcs[];
92 158
93 SkSSE2ProcCoeffXfermode::SkSSE2ProcCoeffXfermode(SkReadBuffer& buffer) 159 SkSSE2ProcCoeffXfermode::SkSSE2ProcCoeffXfermode(SkReadBuffer& buffer)
94 : INHERITED(buffer) { 160 : INHERITED(buffer) {
95 fProcSIMD = reinterpret_cast<void*>(gSSE2XfermodeProcs[this->getMode()]); 161 fProcSIMD = reinterpret_cast<void*>(gSSE2XfermodeProcs[this->getMode()]);
96 } 162 }
(...skipping 136 matching lines...)
233 NULL, // kSrcOut_Mode 299 NULL, // kSrcOut_Mode
234 NULL, // kDstOut_Mode 300 NULL, // kDstOut_Mode
235 NULL, // kSrcATop_Mode 301 NULL, // kSrcATop_Mode
236 NULL, // kDstATop_Mode 302 NULL, // kDstATop_Mode
237 NULL, // kXor_Mode 303 NULL, // kXor_Mode
238 NULL, // kPlus_Mode 304 NULL, // kPlus_Mode
239 NULL, // kModulate_Mode 305 NULL, // kModulate_Mode
240 NULL, // kScreen_Mode 306 NULL, // kScreen_Mode
241 307
242 NULL, // kOverlay_Mode 308 NULL, // kOverlay_Mode
243 NULL, // kDarken_Mode 309 darken_modeproc_SSE2,
244 NULL, // kLighten_Mode 310 lighten_modeproc_SSE2,
245 NULL, // kColorDodge_Mode 311 NULL, // kColorDodge_Mode
246 NULL, // kColorBurn_Mode 312 NULL, // kColorBurn_Mode
247 NULL, // kHardLight_Mode 313 NULL, // kHardLight_Mode
248 NULL, // kSoftLight_Mode 314 NULL, // kSoftLight_Mode
249 NULL, // kDifference_Mode 315 NULL, // kDifference_Mode
250 NULL, // kExclusion_Mode 316 NULL, // kExclusion_Mode
251 multiply_modeproc_SSE2, 317 multiply_modeproc_SSE2,
252 318
253 NULL, // kHue_Mode 319 NULL, // kHue_Mode
254 NULL, // kSaturation_Mode 320 NULL, // kSaturation_Mode
255 NULL, // kColor_Mode 321 NULL, // kColor_Mode
256 NULL, // kLuminosity_Mode 322 NULL, // kLuminosity_Mode
257 }; 323 };
258 324
259 SkProcCoeffXfermode* SkPlatformXfermodeFactory_impl_SSE2(const ProcCoeff& rec, 325 SkProcCoeffXfermode* SkPlatformXfermodeFactory_impl_SSE2(const ProcCoeff& rec,
260 SkXfermode::Mode mode) { 326 SkXfermode::Mode mode) {
261 void* procSIMD = reinterpret_cast<void*>(gSSE2XfermodeProcs[mode]); 327 void* procSIMD = reinterpret_cast<void*>(gSSE2XfermodeProcs[mode]);
262 328
263 if (procSIMD != NULL) { 329 if (procSIMD != NULL) {
264 return SkNEW_ARGS(SkSSE2ProcCoeffXfermode, (rec, mode, procSIMD)); 330 return SkNEW_ARGS(SkSSE2ProcCoeffXfermode, (rec, mode, procSIMD));
265 } 331 }
266 return NULL; 332 return NULL;
267 } 333 }
OLDNEW
« no previous file with comments | « no previous file | no next file » | no next file with comments »

Powered by Google App Engine